Date of Award

Fall 2024

Document Type

Open Access Dissertation

Department

Epidemiology and Biostatistics

First Advisor

Anthony Alberg

Abstract

Background: An efficacious cancer screening test leads to early detection and better prognosis. However, there is currently no efficacious population-based screening test that meets the criteria for screening for ovarian cancer. Serum biomarkers are often affected during bodily injuries including ovarian cancer. The differential behavior of these serum biomarkers could be valuable in the early detection of ovarian cancer.

Objective: The aims of this study were to Aim 1: evaluate the association between individual characteristics and ovarian cancer-related biomarkers in “healthy” women; Aim 2: assess the diagnostic performance of serum biomarkers in detecting ovarian cancer; and Aim 3: examine the diagnostic performance of multiple biomarker combinations in detecting ovarian cancer.

Methods: The study utilized data of women in a nested case-control study designed within the ovarian cancer screening arm of the Prostate Lung Colorectal and Ovarian (PLCO) Cancer Screening Trial. The study consisted of 118 ovarian cancer cases and 951 controls matched to the cases on age in five-year categories and sample year of blood collection. In Aim 1, among the controls, multivariable generalized linear models with log normal distribution using the link function were fitted to examine the association between the women’s characteristics and serum biomarker concentrations. In Aim 2, multivariable logistic regression models were fitted to the entire study population of cases and controls to examine the association between individual biomarkers (n=27) and ovarian cancer. Receiver operating characteristics (ROC) curve analyses were used to compute the sensitivity, specificity and area under the curve (AUC) of each biomarker in detecting ovarian cancer. Their diagnostic performances were also examined by the time of blood draw to cancer diagnosis. In Aim 3, the thirteen biomarkers that were statistically significantly associated with ovarian cancer or had a statistically significant AUC were used to generate multiple biomarker combinations taking into account their biological function. ROC curves were used to assess the diagnostic performance of the multiple biomarker combinations generated. Predictive modelling using random forest and logistic regression with forward selection were used to select the best combinations of biomarkers out of the 27 biomarkers. ROC curves were used to estimate the diagnostic performance of the multiple biomarker combinations generated from the predictive models.

Results: Aim 1: Older age was significantly associated with higher concentrations of HE4, B2M, EGFR, IGFBPII, KLK6, MLN, MMP3, MMP7, OPN, SLPI and SPON2, but lower levels of IGFII and ITIH4. Current smokers had elevated levels of B7-H4 (β=0.50, 95%CI: 0.28, 0.72), CA15.3 (β=0.10, 95%CI: 0.01, 0.19), and OPN (β=11.48, 95%CI: 4.47, 18.49) compared to never smokers. Other characteristics of the women that were found to be significantly associated with some biomarkers included aspirin use, hormone therapy use, race and ethnicity, education status, family history of breast or ovarian cancer and having a comorbidity. Aim 2: Eleven biomarkers, MIF, B7-H4, HE4, CA125, CA15.3, CA72.4, KLK6, mesothelin, MMP7, spondin 2 and prolactin, were significantly elevated in the women with ovarian cancer compared to the women without ovarian cancer with odds ratios ranging from 1.26 – 11.80. The women with cancer had 34% (OR=0.66, 95%CI: 0.45, 0.99) significantly lower levels of eotaxin compared to women without cancer. The top performing biomarkers according to their Youden’s J statistic were CA125, HE4, CA72.4, KLK6 and CA15.3 with sensitivity ranging from 33 – 50% and specificity ranging from 71 – 99%. The biomarkers performed better when examined within 1 year time to ovarian cancer diagnosis compared to more than 1 year time to ovarian cancer diagnosis. Aim 3: The biomarker combination B2M prolactin TRF HE4 MIF CA125 had the best performance (Sensitivity=59%, Specificity=91%; AUC=0.80(0.75, 0.86). The diagnostic performance for the biomarker combination B2M Prolactin TRF HE4 MIF CA125 had 78% sensitivity, 91% specificity and AUC of 0.91(0.87, 0.96) for the women with time to cancer diagnosis of one year and below. The sensitivity, specificity and AUC for the same combination for women diagnosed more than one year time after blood draw were 71%, 63% and 0.72(0.64, 0.79) respectively. In the logistic regression predictive model, which yielded more robust results than the random forest, the best algorithm HE4 CA125 MIF SLPI TRF CA199 generated had sensitivity of 64%, specificity of 87% and AUC of 0.81(0.76, 0.86). When stratified by time to diagnosis, the algorithm HE4 CA125 MIF SLPI TRF CA199 had sensitivity of 76% and a specificity of 90% with AUC of 0.90(0.85, 0.95) for women diagnosed within one year of blood draw compared to sensitivity of 66% and specificity of 75% with AUC of 0.73(0.66, 0.81) in women diagnosed more than one year after the blood draw.

Conclusion: The study found significant variation in some biomarker concentrations by individual level factors in “healthy” women suggesting the need to individualize thresholds of biomarkers to better improve their clinical utility. Despite assessing 27 biomarkers, CA125 and HE4 remain the best performing biomarkers for the detection of ovarian cancer. However, other biomarkers such as KLK6, CA72.4 and MMP7 showed promise. Multiple biomarker combinations showed promise in improving early detection of ovarian cancers with CA125, HE4, TRF and MIF consistently selected as part of the best algorithms. While the results are promising, further validation and exploration of these biomarkers are essential to refine and implement these algorithms in clinical practice effectively.

Rights

© 2025, Maxwell Akonde

Download

Available for download on Thursday, December 31, 2026

Included in

Epidemiology Commons

Share

COinS